Here was a fun thing I recently did; I had to check a php installation for configuration options. So at the command line, I typed
php | tee checkfile.html
< ? phpinfo()?>
ctrl-D
And then examined the contents of checkfile.html (lynx checkfile.html).
A client came to me and asked that I configure her computer with SLI for some high-end gaming she enjoyed. I didn’t know much about SLI but looked around on the web, found that it took two of the same kind of nVidia chip PCIe cards, and headed off to purchase what I needed.
The first time, I came back with two PNY GeForce 6600 DDR PCI-E cards and an ECS Elitegroup 945P-A. As it turns out, just because an Nvidia card is labeled ‘SLI Ready,’ it doesn’t necessarily mean it’ll work out-of-the-box. Your motherboard also has to be SLI ready, and this ECS board isn’t. I could use both cards, this board has two PCIe connections, and since both cards have two monitor outputs, I could have four blazing fast monitors hooked up, but that wasn’t my goal.
I checked the nvidia website and found a list of SLI-compatible motherboards. So I took the board back and got a different one; a Gigabyte i-DNS series nForce 4 SLI Intel Edition board. Note, this board comes with the little bridge piece that crosses over the tops of the two video cards and lets them work together. This SLI bridge piece is the responsibility of the motherboard manufacturer, because the length depends on the distance between the two required PCI Express slots. This bridge does not come from nVidia. I picked up the two cards and went to place them into the Gigabyte motherboard and … realized that these two PNY GeForce 6600, labelled ‘SLI Ready’ did not have a place for this bridge. They would not work either. It is possible, even likely, that other PNY GeForce 6600 boards, the GT for instance would work, but the box I had only had the 6600 on it, and a SLI-Ready logo.
Semi-annoyed, I took the cards back to the shop and brought back two staggeringly heavy PNY GeForce 6800 GS cards. I actually made them open a box in the store to make sure I had ones that would take the little SLI bridge piece. And they did. I brought them home and started setting them up. Everything went together easily and well; and the boards clicked into the gigabyte board a lot easier than the others had gone into the ECS board. On the end of each of the new video cards was a connector, and a dongle attached to them to power these cards. It looked like a normal power tap sort of thing like used for fans and lights, so I plugged in one end, and went to attach the other end … and found that I needed to hook power to both ends of the card; each of these new video cards needed two power connections. I had to return to the store for a bigger, faster, stronger power supply.
With a new power supply in the box, I managed to get it booting. SLI connections were easy; once Windows XP had configured the second video card, it told me that SLI was available and that I should click on a little box to enable it. I did, and my client was thrilled.
Five hours of my time. Five hours beating my head against one stupid problem.
Web (Apache) server had 120 virtual domains on it. 118 of them worked fine; two were not updating their access_log files. I beat the machine, fighting with ulimit values, upgrading the kernel, examining all of the httpd configurations, killing processes that might be using up file descriptors.
Using “lsof -u apache” showed me, finally, that the log files were not in /log/ but were in the webdir: /www/. How they got there was another mystery. Until I found a stack of misconfiguration files in a subdirectory of the virtual hosts definition directory.
Apparently the Apache 2 directive ‘include’ is recursive.
Five hours…
We have a mail server that usually runs around 500 messages in its queue. About a week ago it shot up to 900, then 1200. And then it showed up on the Spamcop blacklist. I lowered the timeouts for how long we kept messages, hoping to flush them out, and lowered the rating at which I sent them to /dev/null (discarding messages where the spam score was over twelve). But that hasn’t helped yet.
So, now I’m installing clamd and mimedefang. Two of my other favorite tools. Unfortunately, it’ll take a hard hit on the system’s load, but hopefully it’ll be worth it.
I recently had the pleasurable honor of reinstalling Fedora Core 4 on a X86-64 system. It wasn’t working very well from my CDs, we kept getting a kernel panic, unable to find a good ‘root’ parameter. I wound up following this scheme, outlaid in the FC 3 Bible:
Followed 'booting installation from hard disk' instructions
found: FC3 Bible p 41
copy vmlinuz and initrd.img from cdrom to /boot/ on sda
renamed them (as copying) to vmlinuz-boot and initrd-boot.img
edited /etc/grub.conf to include menu entry to boot with them.
title Fedora Core Boot Stuffed
root (hd0,0)
kernel /vmlinuz-boot
initrd /initrd-boot.img
rebooted system, went into menu and chose my new config
told it to install from cdrom.
